14 Steven Franks
(2) Nevozmožno [perejti ètot most samomu/*sam].
impossible cross.inf this bridge self.dat/*nom
‘It is impossible to cross this bridge on one’s own.’
(3) Ivan ne znaet [kak tuda dobrat’sja odnomu/*odin].
Ivan not knows how there get.inf alone.dat/*nom
‘Ivan does not know how to get there alone.’
This phenomenon, dubbed by Comrie (1974) the “second dative” (SD), has since
inspired a veritable industry of research.^1 These are the central facts; additional com-
plexities will be introduced in the next section.
A primary issue raised by (1)–(3) is the following: Why does the SD appear in (2)
and (3) but not in (1)? That is, once a mechanism is postulated for assigning the dative
case, the question arises of why that mechanism is not also available even when there
is an accessible antecedent, as in (1). This under-appreciated puzzle, which I call “the
overgeneration problem,” will be my point of departure in this paper. The SD mech-
anism must not be allowed to operate spuriously, since subject OC always induces
obligatory agreement. The problem of avoiding overgeneration of the SD is partic-
ularly recalcitrant in current minimalist (and other) models, which build syntactic
structures from the bottom up and apply syntactic operations in a strictly local fash-
ion. This architecture introduces two related timing issues, at least, if case is assigned
or valued on-line (i.e. in a derivational and/or cyclic fashion). First, as noted, given a
local mechanism to assign the dative, it becomes difficult to block the SD just in case
an accessible antecedent is ultimately going to be introduced into the structure. Sec-
ond, it is unclear how to determine the case to be assigned before that ultimate acces-
sible antecedent is merged, which could potentially be an unlimited distance from the
semipredicative:
(4) My rešili [postarat’sja [delat’ èto sami/*samim]].
we.nom decided try.inf do.inf this self.pl.nom/*dat
‘We decided to try to do this ourselves.’
Resolving these issues would seem to require look-ahead, that is, consultation of infor-
mation not yet introduced into the derivation. There are two general kinds of solution
to such look-ahead. One is checking, which involves selection of fully formed lexi-
cal items, complete with all features. This may involve postponing validation of their
properties until LF. The other is to value features (i.e. to assign feature values) once- Representative publications that deal with the SD include: Neidle (1988), Franks
(1990, 1995 ), Greenberg & Franks (1991), Babby (1998, 2009 ), Moore & Perlmutter (2000),
Sigurðsson (2002), Madariaga (2006), and Landau (2008). For a recent brief overview, see
Bailyn (2012: §5.1.4.1).